rust-par2 0.1.2

Pure Rust PAR2 verify and repair with SIMD-accelerated Galois field arithmetic
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234

//! Micro-benchmarks to identify hotspots in the GF multiply-accumulate pipeline.
//!
//! Usage: cargo run --release --example microbench
//!
//! Measures:
//!   1. XOR throughput (memory bandwidth ceiling)
//!   2. Single-source GF(2^16) mul_add throughput
//!   3. Multi-source accumulate (simulates repair inner loop)
//!   4. Repair simulation with realistic parameters
//!
//! Reports GB/s and GF-muls/sec so you can compare against theoretical peak.

use std::time::{Duration, Instant};

fn main() {
    println!("=== rust-par2 micro-benchmarks ===\n");

    // Detect CPU features
    #[cfg(target_arch = "x86_64")]
    {
        println!("CPU features:");
        println!("  AVX2:  {}", is_x86_feature_detected!("avx2"));
        println!("  SSSE3: {}", is_x86_feature_detected!("ssse3"));
        println!("  AVX-512F: {}", is_x86_feature_detected!("avx512f"));
        println!("  PCLMULQDQ: {}", is_x86_feature_detected!("pclmulqdq"));
        println!();
    }

    let cores = std::thread::available_parallelism()
        .map(|n| n.get())
        .unwrap_or(1);
    println!("Available parallelism: {cores} threads\n");

    // Warm up GF tables
    rust_par2::gf_simd_public::mul_add_buffer(&mut [0u8; 4], &[0u8; 4], 1);

    // --- Benchmark parameters ---
    let slice_size = 768_000; // PAR2 default block size (768 KB)
    let large_buf = 4 * 1024 * 1024; // 4 MB for throughput tests

    bench_xor_throughput(large_buf);
    bench_single_mul_add(large_buf);
    bench_single_mul_add(slice_size);
    bench_multi_source_accumulate(slice_size, 16);
    bench_multi_source_accumulate(slice_size, 64);
    bench_multi_source_accumulate(slice_size, 128);
    bench_multi_source_accumulate(slice_size, 512);
    bench_multi_source_accumulate(slice_size, 1365); // ~1GB / 768KB

    println!("\n--- Repair simulation (parallel, rayon) ---\n");
    bench_repair_simulation(slice_size, 1365, 40); // 1GB data, 3% = ~40 damaged
    bench_repair_simulation(slice_size, 1365, 10); // 1GB data, ~0.7% damaged
}

/// Pure XOR — establishes the memory bandwidth ceiling.
fn bench_xor_throughput(buf_size: usize) {
    let src = vec![0xAAu8; buf_size];
    let mut dst = vec![0x55u8; buf_size];

    let (iters, elapsed) = bench_loop(Duration::from_secs(2), || {
        rust_par2::gf_simd_public::xor_buffers(&mut dst, &src);
    });

    let total_bytes = buf_size as f64 * iters as f64;
    let gbps = total_bytes / elapsed.as_secs_f64() / 1e9;
    // Each iteration reads src + reads dst + writes dst = 3 × buf_size
    let mem_gbps = 3.0 * total_bytes / elapsed.as_secs_f64() / 1e9;
    println!("XOR throughput ({}):", fmt_size(buf_size));
    println!("  {gbps:.2} GB/s data, {mem_gbps:.2} GB/s memory traffic");
    println!("  {iters} iterations in {:.3}s", elapsed.as_secs_f64());
    println!();
}

/// Single constant × buffer — measures raw SIMD GF multiply speed.
fn bench_single_mul_add(buf_size: usize) {
    let src = random_buf(buf_size);
    let mut dst = vec![0u8; buf_size];
    let constant = 12345u16;

    let (iters, elapsed) = bench_loop(Duration::from_secs(2), || {
        rust_par2::gf_simd_public::mul_add_buffer(&mut dst, &src, constant);
    });

    let total_bytes = buf_size as f64 * iters as f64;
    let gbps = total_bytes / elapsed.as_secs_f64() / 1e9;
    let gf_muls = (buf_size / 2) as f64 * iters as f64; // u16 elements
    let gf_muls_per_sec = gf_muls / elapsed.as_secs_f64();
    // Memory: read src + read dst + write dst = 3 × buf_size
    let mem_gbps = 3.0 * total_bytes / elapsed.as_secs_f64() / 1e9;
    println!("Single mul_add ({}):", fmt_size(buf_size));
    println!("  {gbps:.2} GB/s data, {mem_gbps:.2} GB/s memory traffic");
    println!("  {:.2} billion GF-muls/sec", gf_muls_per_sec / 1e9);
    println!("  {iters} iterations in {:.3}s", elapsed.as_secs_f64());
    println!();
}

/// Multi-source accumulate: dst ^= Σ c[i] * src[i] — the repair inner loop.
fn bench_multi_source_accumulate(slice_size: usize, num_sources: usize) {
    let sources: Vec<Vec<u8>> = (0..num_sources).map(|_| random_buf(slice_size)).collect();
    let coeffs: Vec<u16> = (1..=num_sources as u16).collect();
    let mut dst = vec![0u8; slice_size];

    // Sequential approach (what repair currently does per damaged block)
    let (iters, elapsed) = bench_loop(Duration::from_secs(2), || {
        dst.fill(0);
        for (src, &coeff) in sources.iter().zip(coeffs.iter()) {
            rust_par2::gf_simd_public::mul_add_buffer(&mut dst, src, coeff);
        }
    });

    let total_data = slice_size as f64 * num_sources as f64 * iters as f64;
    let gbps = total_data / elapsed.as_secs_f64() / 1e9;
    let gf_muls = (slice_size / 2) as f64 * num_sources as f64 * iters as f64;
    let gf_muls_per_sec = gf_muls / elapsed.as_secs_f64();
    // Memory per iteration: num_sources × (read src + read dst + write dst)
    // But dst stays hot in cache after first source, so realistically:
    //   num_sources × read_src + ~2 × slice_size (dst read/write, cached)
    let mem_cold =
        (num_sources as f64 * slice_size as f64 + 2.0 * slice_size as f64) * iters as f64;
    let mem_gbps = mem_cold / elapsed.as_secs_f64() / 1e9;
    let per_iter_ms = elapsed.as_secs_f64() * 1000.0 / iters as f64;

    println!(
        "Multi-source accumulate ({} srcs × {}):",
        num_sources,
        fmt_size(slice_size)
    );
    println!("  {gbps:.2} GB/s src data, ~{mem_gbps:.2} GB/s estimated mem traffic");
    println!("  {:.2} billion GF-muls/sec", gf_muls_per_sec / 1e9);
    println!(
        "  {per_iter_ms:.3} ms per accumulation ({iters} iters in {:.3}s)",
        elapsed.as_secs_f64()
    );
    println!();
}

/// Simulates the full repair inner loop with rayon parallelism.
/// num_sources total blocks, num_damaged blocks to reconstruct.
fn bench_repair_simulation(slice_size: usize, num_sources: usize, num_damaged: usize) {
    use rayon::prelude::*;

    let sources: Vec<Vec<u8>> = (0..num_sources).map(|_| random_buf(slice_size)).collect();
    // Build a coefficient matrix [num_damaged × num_sources]
    let coeffs: Vec<Vec<u16>> = (0..num_damaged)
        .map(|d| {
            (0..num_sources)
                .map(|s| ((d * 31 + s * 17 + 1) % 65535) as u16)
                .collect()
        })
        .collect();

    let src_refs: Vec<&[u8]> = sources.iter().map(|s| s.as_slice()).collect();

    // Time the parallel repair loop
    let t = Instant::now();
    let _results: Vec<Vec<u8>> = (0..num_damaged)
        .into_par_iter()
        .map(|dmg_i| {
            let mut dst = vec![0u8; slice_size];
            for (src_idx, src_data) in src_refs.iter().enumerate() {
                let coeff = coeffs[dmg_i][src_idx];
                if coeff != 0 {
                    rust_par2::gf_simd_public::mul_add_buffer(&mut dst, src_data, coeff);
                }
            }
            dst
        })
        .collect();
    let elapsed = t.elapsed();

    let total_src_data = slice_size as f64 * num_sources as f64 * num_damaged as f64;
    let gbps = total_src_data / elapsed.as_secs_f64() / 1e9;
    let gf_muls = (slice_size / 2) as f64 * num_sources as f64 * num_damaged as f64;
    let total_data_gb = num_sources as f64 * slice_size as f64 / 1e9;

    println!(
        "Repair sim: {} srcs × {} damaged × {} = {:.1} GB total data",
        num_sources,
        num_damaged,
        fmt_size(slice_size),
        total_data_gb,
    );
    println!("  {:.3}s elapsed", elapsed.as_secs_f64());
    println!("  {gbps:.2} GB/s src throughput (parallel)");
    println!(
        "  {:.2} billion GF-muls/sec (parallel)",
        gf_muls / elapsed.as_secs_f64() / 1e9
    );
    println!(
        "  {:.1} ms per damaged block",
        elapsed.as_secs_f64() * 1000.0 / num_damaged as f64,
    );
    println!();
}

// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------

/// Run `f` repeatedly for at least `min_duration`, return (iterations, elapsed).
fn bench_loop<F: FnMut()>(min_duration: Duration, mut f: F) -> (u64, Duration) {
    // Warmup
    for _ in 0..3 {
        f();
    }

    let mut iters = 0u64;
    let start = Instant::now();
    loop {
        f();
        iters += 1;
        let elapsed = start.elapsed();
        if elapsed >= min_duration && iters >= 4 {
            return (iters, elapsed);
        }
    }
}

fn random_buf(size: usize) -> Vec<u8> {
    use rand::RngCore;
    let mut buf = vec![0u8; size];
    rand::rng().fill_bytes(&mut buf);
    buf
}

fn fmt_size(bytes: usize) -> String {
    if bytes >= 1024 * 1024 {
        format!("{:.1} MB", bytes as f64 / (1024.0 * 1024.0))
    } else if bytes >= 1024 {
        format!("{} KB", bytes / 1024)
    } else {
        format!("{bytes} B")
    }
}